training-infrastructure

Here are 4 public repositories matching this topic...

Brain2nd / NeuronSpark-Trainer

Pre-training for NeuronSpark: streaming data lines from the open MiniCPM corpora, whole-step CUDA graphs, exact resume of data, optimizer and W&B, trunk and decay/long-context branches, and an unattended supervisor that shares the GPU with other jobs. Trains a 2B model at ~48k tokens/s on one GB300.

pytorch spiking-neural-networks language-model data-pipeline checkpointing pretraining wandb llm-training minicpm nvidia-blackwell cuda-graphs training-infrastructure

Updated Nov 2, 2026
Python

Tianhao1314 / slurm-resilient-training

Star

Zero-loss deep learning training on shared SLURM clusters: relay chains, durable checkpoints, single-writer guards, queue economics

bash deep-learning hpc fault-tolerance slurm pytorch gpu-cluster checkpointing training-infrastructure

Updated Aug 16, 2026
Shell

huzjie / beamforge

Star

Sparse frontier-model training factory (Reflection Beam direction)

python reinforcement-learning deep-learning fault-tolerance sandbox moe zero-dependency distributed-training mixture-of-experts llm grpo training-infrastructure

Updated Oct 6, 2026
Python

hxm2023 / GRPO-Guard

Star

Machine-verifiable evidence chain for online LLM post-training (TRL GRPO + vLLM): trajectory contracts, fault injection F1-F8, guarded updates, paired gradient replay

fault-injection trl rlhf vllm llm-observability grpo llm-post-training training-infrastructure

Updated Sep 29, 2026
Python

Add this topic to your repo

To associate your repository with the training-infrastructure topic, visit your repo's landing page and select "manage topics."

Learn more